Personalized online video recommendation by neighborhood score propagation based global ranking

  • Authors:

  • Chunxi Liu;Shuqiang Jiang;Qingming Huang

  • Affiliations:

  • Graduate University of Chinese Academy of Sciences, Beijing, China;Chinese Academy of Sciences, Beijing, China;Graduate University of Chinese Academy of Sciences, Beijing, China and Chinese Academy of Sciences, Beijing, China

  • Venue:
  • Proceedings of the First International Conference on Internet Multimedia Computing and Service
  • Year:
  • 2009

Quantified Score

Hi-index 0.00

Visualization

Abstract

With the fast rising of the video sharing website, such as Youtube etc, there is an emergent requirement to provide video recommendation service according to user's interest. However, most of the current video recommendation systems are based on text analysis and utilize local ranking algorithms. Text based approaches may fail when the textual information is incomplete. The performance of the local ranking algorithm is limited by the fact that it only considers the relation between the target item and the un-target items, and neglects the useful information among the un-target items. In this paper, we propose a novel personalized framework to achieve recommendation by re-ranking the video search result list according to user selected one by using multimodal features. For re-ranking, the neighborhood score propagation based global ranking approach is adopted. This algorithm explores the inner structures of the video data distribution and ensures that similar videos have similar recommendation scores. In our approach, firstly the video search result list is obtained according to user's specified query through video search engine. Then, multimodal features, including aural, textual and visual features, are extracted from the user clicked video and the videos in the search result list respectively. In this step, we propose to use the concept probability distribution feature to represent the video visual content. This feature reveals the concepts consisting of the video and is suitable for video high level similarity representation. After that, the multi-graph learning framework is explored to re-rank these videos. In summary, our approach could provide personalized video recommendation service according to user's demand through an interactive video re-ranking. The experimental results and evaluations show that the proposed approach is effective.